This is fixing post-FIN state (by disabling offload post-FIN).
Once the tcp state enters half-closing tcp state (fin wait,
close wait), delete the offload rules.
From this commit, we have done the short term solution for conntrack
tcp timeout issue. Here is what we have done so far.
- Stop updating tcp timeout to avoid updating wrong tcp state
timeout.
- Update the ESTABLISHED timeout
nf_conntrack_tcp_timeout_established (432000) after adding
bidirectional rules.
- Delete the tcp rules when the tcp state has left
"established".
Here is the long term solution and need to be addressed in follow up
commits.
- Parse the tcp state from netlink conntrack event.
- Build a mapping to trace the tcp state of the tcp conntrack event.
- Update tcp state timeout for {ESTABLISHED (432000), FIN_WAIT (120),
CLOSE_WAIT (60)}.
Bug: 190783768
Bug: 192804833
Test: atest TetheringCoverageTests
Manual test:
1. Browse on tethered device
2. Check conntrack tcp state is established.
$ cat /proc/net/nf_conntrack
ipv4 2 tcp 6 431995 ESTABLISHED src=192.168.207.9
dst=140.112.8.116 sport=50697 dport=443 ..
3. Check offload rules are added via dumpsys.
Upstream:
tcp .. 192.168.207.9:50697 -> 14(rmnet0) 10.224.1.247:50697
-> 140.112.8.116:443 ..
Downstream:
tcp .. 140.112.8.116:443 -> 30(30) 10.224.1.247:50697
-> 192.168.207.9:50697 ..
4. Stop browsing for a few seconds.
5. Check conntrack tcp state is half-closed.
$ cat /proc/net/nf_conntrack
ipv4 2 tcp 6 116 TIME_WAIT src=192.168.207.9
dst=140.112.8.116 sport=50697 dport=443 ..
5. Check offload rules are removed via dumpsys.
Upstream:
(not found)
Downstream:
(not found)
Change-Id: I07e27230bf8952acd7828d1f605167758b3bc490
Needed because a payload data packet may have gone through
non-offload path, before we added offload rules, and that this
may result in in-kernel conntrack state being in ESTABLISHED
but pending ACK (ie. UNACKED) state. But the in-kernel conntrack
might never see the ACK because we just added offload rules.
As such after adding the rules we need to force the timeout back
to the normal ESTABLISHED timeout of 5 days.
Issue: the timeout is set to unacknowledged 300s (countdwon to 298s)
$ adb shell cat /proc/net/nf_conntrack
ipv4 2 tcp 6 298 ESTABLISHED src=192.168.244.128
dst=140.112.8.116 sport=45694 dport=443 ..
Test: atest TetheringCoverageTests
Manual check:
$ adb shell cat /proc/net/nf_conntrack
ipv4 2 tcp 6 431988 ESTABLISHED src=192.168.40.162
dst=140.112.8.116 sport=40774 dport=443 ..
Bug: 190783768
Bug: 192804833
Change-Id: I8c34e85e26c9d976e5e2b85473db75ff46d8abd4
This is a preparation for only update the tcp timeout while
adding rules. Also add slack time for updating UDP timeout
interval.
Bug: 190783768
Bug: 192804833
Test: atest TetheringCoverageTests
Change-Id: I3151b531e6581e257f3cfa39ad2fcf1650358b3d
This ensures classes are used from the service-connectivity jar, instead
of using classes from the system_server bootclasspath when there is a
name conflict.
Any developer adding a future class should do so in a subpackage of
com.android.connectivity (such as com.android.connectivity.server).
Otherwise, jarjar rules need to be added manually until b/180995093 is
fixed.
Also update current jarjar rules so that classes are jarjared to
com.android.connectivity.[original name], making it easier to find the
original source. This is consistent with the wifi module.
Bug: 193086215
Test: atest CtsNetTestCases
dexdump on service-connectivity.jar shows no classes outside of
com.android.connectivity and com.android.server
Original-Change: https://android-review.googlesource.com/1759589
Merged-In: I2aadeca32751267b74d4fd2fd93bb3e8c62e46c0
Change-Id: I2aadeca32751267b74d4fd2fd93bb3e8c62e46c0
Currently netd supports only the default value for VPN but CS
send priorty value 1 to netd. It will break the default routing
for VPN. Thus, update network preference priority value to 0 for
VPN.
Bug: 193245476
Test: atest CtsHostsideNetworkTests:HostsideVpnTests
Original-Change: https://android-review.googlesource.com/1762867
Merged-In: I197cb358e8e30355fbf675e4c623abebe7abdb7f
Change-Id: I197cb358e8e30355fbf675e4c623abebe7abdb7f
This ensures classes are used from the service-connectivity jar, instead
of using classes from the system_server bootclasspath when there is a
name conflict.
Any developer adding a future class should do so in a subpackage of
com.android.connectivity (such as com.android.connectivity.server).
Otherwise, jarjar rules need to be added manually until b/180995093 is
fixed.
Also update current jarjar rules so that classes are jarjared to
com.android.connectivity.[original name], making it easier to find the
original source. This is consistent with the wifi module.
Bug: 193086215
Test: atest CtsNetTestCases
dexdump on service-connectivity.jar shows no classes outside of
com.android.connectivity and com.android.server
Change-Id: I2aadeca32751267b74d4fd2fd93bb3e8c62e46c0
Currently netd supports only the default value for VPN but CS
send priorty value 1 to netd. It will break the default routing
for VPN. Thus, update network preference priority value to 0 for
VPN.
Bug: 193245476
Test: atest CtsHostsideNetworkTests:HostsideVpnTests
Change-Id: I197cb358e8e30355fbf675e4c623abebe7abdb7f
